Supervising an autonomous coding agent can feel less like programming and more like a hostage negotiation: something you value is in the hands of a party that acts on it directly, while you can only influence how it acts. Taken as a structural allegory rather than a claim about machine intent, the frame explains a frustration that prompt and context engineering leave untouched and names a skill the field has not yet named: setting terms the agent will hold, reading what it does against what it says, calibrating trust turn by turn, and steering a volatile collaboration toward a result you can live with.
Zenodo ยท Jul 2026